Consults on WHICH open-source UI components to use, HOW to remap them to your design tokens, and WHERE to place them — so an AI-built frontend is coherent with the site's message, true to its DESIGN.md, and lands in the distinctive/awwwards band instead of bland or average-AI. Use when the user asks "what components should I use", "컴포넌트 골라줘 / 추천해줘", "pick UI components", "which library for this section", or right after a DESIGN.md / design-token file exists and before building. Live-scans the component galleries (WebFetch/WebSearch) for current components, with a model-knowledge fallback. Works for live frontends AND HTML-slide decks.
Create polished, image-embedded HTML presentation decks (1920×1080 HTML PPT / 카드뉴스 / keynote) from a topic using the "raw5" method — exactly 5 AI-generated background images bound to CSS variables and reused across every slide via wash / blend / tone / grid / mask / material-image-card techniques, while all text, numbers, and charts stay as crisp HTML/SVG. Runs a strict 3-stage workflow: deck plan → 5 image prompts (HARD STOP so the user generates them in gpt-image-2 or another tool) → HTML build. Use this whenever the user wants an HTML slide deck, an HTML 발표자료, a 카드뉴스, an image-heavy/visual presentation, a "raw5" or "iamht" deck, or hands you 5 images and asks for slides — even if they don't name the technique. Four design modes: V7 bright report, V8 dark brutalist, university AX grid, street editorial.